design

name	design
description	Professional design system - illustrations, UI mockups, social graphics, brochures, logos, infographics, and image editing. Routes to optimal pipeline (HTML Studio for layouts, Replicate AI for creative imagery, PPTX for presentations). Use when user asks to design, create visuals, generate graphics, make a mockup, create a brochure, design a logo, or produce any visual asset. Do NOT use for simple one-off image generation (use /flux-image instead).
argument-hint	[type] [brief]
allowed-tools	Read, Write, Edit, Bash(python3:*)
model	sonnet
metadata	{"author":"Misha Hanin","email":"[e-mail]","version":"1.0"}
x-heading-orchestration	{"parallel_safe":true,"shared_state":[],"triggers":["design social","design infographic","design mockup","design illustration","design logo"]}
x-heading-capability	{"what":"Professional design studio for social graphics, infographics, UI mockups, cards, brochures, logos, and AI imagery. Routes each task to the optimal pipeline - HTML Studio (free, layout/text), Replicate AI (creative imagery, vectors), or PPTX (presentations).\n","how":"Run /design <type> <brief> (e.g. /design infographic, /design logo, /design photo, /design edit). HTML renders land in outputs/design/; AI-generated images in outputs/content/images/.\n","when":"Use for any branded visual asset. For a simple one-off image from a prompt use /flux-image; for presentation slides use /pptx-generator.\n"}
x-heading-routing	{"category":"Design","triggers":["design social","design infographic","design mockup","design illustration","design logo"],"exclusions":["Presentations -> /pptx-generator","image from prompt -> /flux-image"],"compound":"No","router":"auto"}

Design System

Professional design studio with three rendering pipelines: HTML Studio (free, deterministic), Replicate AI (17 models), and Document Production (PPTX, PDF). Routes each task to the optimal pipeline automatically.

Commands

Command	Pipeline	Description
/design social [brief]	HTML Studio	Social media graphic (LinkedIn, Twitter)
/design infographic [brief]	HTML Studio	Data visualization, infographic
/design mockup [brief]	HTML Studio	UI/UX mockup, app screen, dashboard
/design card [brief]	HTML Studio	Branded card, quote card, stat card
/design illustration [brief]	Replicate AI	Creative illustration, conceptual art
/design photo [brief]	Replicate AI	Photorealistic image, product shot
/design logo [brief]	Replicate AI (SVG)	Logo, icon, native vector graphic
/design text-graphic [brief]	Replicate AI	Graphic with embedded text (poster, banner)
/design brochure [brief]	HTML + Replicate	Layout from HTML, illustrations from AI
/design edit [instruction]	Replicate AI	Edit existing image via natural language
/design upscale [path]	Replicate AI	Upscale image for print quality
/design remove-bg [path]	Replicate AI	Remove background from image
/design draft [brief]	Replicate AI	Fast concept at $0.003 via FLUX Schnell

Routing Logic

Pipeline 1: HTML Studio (free, $0)

When: Task is layout-driven, text-driven, or data-driven.

Routes: social, infographic, mockup, card, brochure layouts.

Workflow:

1. Generate a complete, self-contained HTML file with inline CSS
2. Read references/brand.css for the 31C design system variables
3. Apply brand design: dark backgrounds (#000000), Inter font, accent colors (#5B5FFF blue-purple, #FF8C00 orange)
4. Save HTML to outputs/design/source/ for version control
5. Run the renderer:

python3 scripts/design-studio.py render \
  --file outputs/design/source/{name}.html \
  --width {W} --height {H} \
  --brand 31c \
  -o outputs/design/{name}.png

6. Read the generated image to show the user
7. If user wants changes, edit the HTML and re-render

HTML generation rules:

• Self-contained: all CSS inline, no external dependencies except Google Fonts
• Use CSS custom properties from brand.css: var(--accent), var(--bg), var(--text), etc.
• Use utility classes from brand.css: .card, .container, .grid-2, .accent, .display
• Dark backgrounds ONLY - never white or light
• Maximum 6 lines of text per visual
• 40% empty space (breathing room)
• One main message per visual

Pipeline 2: Replicate AI (API-driven)

When: Task needs creative imagery, photography, vectors, or text rendered into images.

Routes: illustration, photo, logo, text-graphic, draft, edit, upscale, remove-bg.

Model selection:

Task	Model	Command
Illustration, brand art	recraft-v4	generate --model recraft-v4
Photorealistic image	flux-2-pro	generate --model flux-2-pro
Logo / icon (vector SVG)	recraft-v4-svg	generate --model recraft-v4-svg
Graphic with text	ideogram-v3	generate --model ideogram-v3
Fast concept draft	flux-schnell	generate --model flux-schnell
Edit existing image	kontext	edit --model kontext --image {path}
Upscale for print	crisp-upscale	upscale --image {path}
Remove background	eraser	remove-bg --image {path}

Generation command:

python3 scripts/design-engine.py generate \
  --model {model} \
  --prompt "{prompt}" \
  --width {W} --height {H} \
  -o outputs/content/images/{name}.png

Edit/upscale/remove-bg:

python3 scripts/design-engine.py edit --model kontext --image {path} --prompt "{instruction}" -o {output}
python3 scripts/design-engine.py upscale --image {path} -o {output}
python3 scripts/design-engine.py remove-bg --image {path} -o {output}

Brand-aware prompting for AI generation: When generating images for 31C, append to prompt: "Dark background, professional, sovereign, bold. Color palette: deep blues, blue-purple (#5B5FFF), orange (#FF8C00) accents. Clean, modern, authoritative."

Pipeline 3: Document Production

When: Task is a presentation deck.

Redirect: "For presentation slides, use /pptx-generator directly - it has the full 16-layout system with 31C branding."

For PDF from HTML, use the design studio:

python3 scripts/design-studio.py pdf --file {html_path} --brand 31c -o {output}.pdf

Standard Dimensions

Format	Width	Height	Use Case
LinkedIn post	1200	628	Standard LinkedIn image
LinkedIn square	1080	1080	Carousel slide
Instagram	1080	1080	Square social
Instagram story	1080	1920	Vertical story
Twitter/X	1200	675	Twitter image
Presentation	1920	1080	16:9 slide
A4 portrait	2480	3508	Print (300 DPI)
A4 landscape	3508	2480	Print landscape

Output Locations

• HTML Studio renders: outputs/design/{name}.png + source outputs/design/source/{name}.html
• AI-generated images: outputs/content/images/{name}.png (consistent with /flux-image)
• Project work (brochures, campaigns): outputs/design/{project-name}/
• Multi-format exports: outputs/design/{name}-{W}x{H}.png

Model Reference

Full model registry with capabilities, pricing, and prompt tips: references/models.md

Brand CSS Reference

Injectable 31C design system (Inter font, dark theme, accent colors, utility classes): references/brand.css

Source of truth for colors: .claude/skills/pptx-generator/brands/31c/brand.json

Voice and Style

• Hyphens only (never em-dashes)
• ODUN.ONE (never ODUN or Odun)
• DPI+ (never DPI Plus)
• Dark backgrounds always (31C is dark-mode only)
• Signal over noise - every element earns its place
• Sovereign and bold - communicate independence and authority

Sensitive sessions (SENSITIVE_MODE)

Before sending any prompt or query to an external API (Replicate imagery, web search/fetch for references), check session sensitivity:

python3 -c "from scripts.utils.sensitive import is_sensitive, sanitize_prompt_guidance; print(sanitize_prompt_guidance()) if is_sensitive() else None"

If it prints guidance (sensitivity not explicitly cleared — the fail-closed default), strip ALL project-identifying detail from the prompt/query before the call: no codenames, company names, people names, deal terms, or strategic specifics. Use abstract, generic descriptions. The asset stays local; only the query goes external. This replaces the removed secure-design vault wrapper.

NEVER

• Never use MCP servers for design work
• Never call external APIs besides Replicate
• Never hardcode API tokens in any file
• Never use light or white backgrounds (31C brand is dark-mode only)
• Never skip brand CSS injection for HTML Studio designs
• Never generate final images without confirming the brief first
• Never output more than needed - start with one variant, iterate